delay and capacity trade-offs in mobile ad hoc networks: a global perspective
DESCRIPTION
Delay and Capacity Trade-offs in Mobile Ad Hoc Networks: A Global Perspective. Gaurav Sharma,Ravi Mazumdar,Ness Shroff IEEE/ACM Transaction on Networking, Vol 15,No 5. 2007, pp981-991. d96725002 蕭 鉢 d96725011 黃文莉 r96725035 林意婷 指導老師 : 林永松 教授. - PowerPoint PPT PresentationTRANSCRIPT
1111IMproves Yourself
Delay and Capacity Trade-offs in Mobile Ad Hoc Networks: A Global Perspective
Delay and Capacity Trade-offs in Mobile Ad Hoc Networks: A Global Perspective
d96725002 d96725002 蕭 鉢蕭 鉢d96725011 d96725011 黃文莉黃文莉r96725035 r96725035 林意婷林意婷
指導老師指導老師 : : 林永松 教授林永松 教授
d96725002 d96725002 蕭 鉢蕭 鉢d96725011 d96725011 黃文莉黃文莉r96725035 r96725035 林意婷林意婷
指導老師指導老師 : : 林永松 教授林永松 教授
Gaurav Sharma,Ravi Mazumdar,Ness ShroffIEEE/ACM Transaction on Networking, Vol 15,No 5. 2007, pp981-991
2222IMproves Yourself
Is node mobility a “liability” or an “asset” in ad hoc networks?
3333IMproves Yourself
Liability Hand-off protocols for cellular networks [Toh & Akyol] Adverse effect on the performance of traditional ad hoc routing
protocols [Bai, Sadagopan and Helmy]
Asset Grossglauser and Tse showed node mobility can increase the
capacity of an ad hoc network, if properly exploited. The delay related issues were not considered.
4444IMproves Yourself
To Provide better understanding of
the delay and capacity trade-offs
in mobile ad hoc networks (MANET)
from a global perspective
5555IMproves Yourself
OutlinesOutlines
Introduction Capacity scaling of ad hoc networks Mobility can increase capacity Main contributions Main Results Overview
The Models Hybrid random walk model i.i.d mobility model Random walk model Hybrid random direction model Discrete random direction model Brownian motion mobility
Critical Delay and 2-hops Delay Critical Delay and 2-hops Delay Under Various Mobility Models
Lower bound on critical delay for hybrid random walk models Upper bound on critical delay for hybrid random walk models Lower bound on critical delay for discrete random direction models Upper bound on critical delay for discrete random direction models
Discussion Characteristic path length
Conclusion
6666IMproves Yourself
IntroductionIntroduction
7777IMproves Yourself
Capacity scaling of ad hoc networksCapacity scaling of ad hoc networks
Study fundamental properties of large wireless networks [Gupta & Kumar]
Derive asymptotic bounds for throughput capacity To derive upper bounds, use:
Interference penalty—nodes within range need to be silenced for successful communication
Multi-hop relaying penalty— a node that traverses a distance of d needs to use order of d hops.
To derive constructive lower bounds, use: Geographic routing strategic along great circles Greedy coloring schedules.
8888IMproves Yourself
Capacity scaling of ad hoc networksCapacity scaling of ad hoc networks
Number of nodes
Per-node Capacity
[Gupta & Kumar]-Nodes static-Interference model: protocol or physical model-Common power level across network
)log
1(
nn
[Francheschetti &Dousse] - Nodes static - Power control allowed
)1
(n
[Grossglauser & Tse] - Nodes are mobile
)1(
9999IMproves Yourself
Mobility can increase capacityMobility can increase capacity
[Grossglauser & Tse] achieve constant capacity scaling by two-hop relaying
[Gupta & Kumar] allow for constant capacity scaling if the traffic pattern is purely local.
Source uses one of all possible mobile nodes as a relay. Source splits stream uniformly across all relays. When a mobile forwarder nears the destination, it hands off packet.
10101010IMproves Yourself
Mobility can increase capacityMobility can increase capacity
Why does mobility increase capacity? By choosing a random intermediate relay, the traffic is diffused uniformly
throughout the network. Thus, on average, every mobile node has a packet for every other destination
and can schedule a packet to a nearby destination in every slot. (For those who took randomized algorithms, this is akin to permutation
routing algorithms)
Catch: forwarding strategy improves capacity at the expense of introducing delay.
Need to study the delay-capacity tradeoff!!
11111111IMproves Yourself
Main contributionsMain contributions
Delay-capacity tradeoff: increasing the maximum allowable average delay increases the capacity.
Delay-capacity tradeoff depends on network setting, mobility patterns.
Different mobility models have been studied in the literature i.i.d Brownian motion Random way-point Random walk
Difficult to compare results across paper because network setting are quite different.
How does the mobility model affect the delay capacity trade-off?
12121212IMproves Yourself
Main Results: Notion of critical delay to compare mobility modes
Main Results: Notion of critical delay to compare mobility modes
For each mobility model, there is a critical delay below which node mobility cannot be exploited for improving capacity.
Critical delay depends mainly on mobility pattern, not on network setting
13131313IMproves Yourself
OverviewOverview
Mobility can increase capacity. Delay-capacity tradeoff depends on network setting, mobility
models. Some questions arises
How representative are these mobility models in this study? Can the delay-capacity relationship be significantly different under the
mobility models? What sort of delay-capacity trade-off are we likely to see in real world
scenario?
14141414IMproves Yourself
Main Results A new hybrid random walk model
Main Results A new hybrid random walk model
Propose and study a new family on hybrid random walk models, indexed by a parameter in [0, ].
For the hybrid random walk model with parameter ,critical delay is
As approaches 0, the hybrid random walk model approaches an i.i.d mobility model.
As approaches , the hybrid random walk model approaches a random walk mobility model.
2
1
)( 2n
21
15151515IMproves Yourself
Main Results A new hybrid random walk model
Main Results A new hybrid random walk model
Number of nodes
Critical Delay
i.i.d )1(
random walk model
)(n
0
1
2
1
Hybrid random walk model )( 2n
16161616IMproves Yourself
Main Results A new hybrid random direction model
Main Results A new hybrid random direction model
Propose and study a new family on hybrid random direction models, indexed by a parameter in [0, ].
For the hybrid random direction model with parameter , the critical delay is
As approaches 0, this hybrid random direction model approaches a random way-point model.
As approaches , this hybrid random direction model approaches a Brownian mobility model.
21
)( 2
1 n
21
17171717IMproves Yourself
Main Results A new hybrid random direction model
Main Results A new hybrid random direction model
Number of nodes
Critical Delay
Brownian mobility
)(n
1
Hybrid random direction model
)( 21 n
Random Way-point
)( n
0
21
18181818IMproves Yourself
The ModelsThe Models
19191919IMproves Yourself
Hybrid random walk modelHybrid random walk model
Divide the unit square into cells of area Divide each cell into sub cells of area In each time slot, a node is in one of sub cells in a cell. At the beginning of a slot, node jumps uniformly to one of the sub
cells of an “Adjacent cell”
2n 2
1
n21n n
1
nn
20202020IMproves Yourself
i.i.d mobility modeli.i.d mobility model
As approaches 0, we get i.i.d mobility. One big cell with n sub-cells. In each slot, a node is in one of the sub-cells. At the beginning of a time slot, a node jumps uniformly to one of
the n subcells.
21212121IMproves Yourself
Random walk modelRandom walk model
As approaches , we get the random walk. n cells, one sub-cell in each cell. In any slot, a node is in particular cell. At the beginning of a slot, node jumps uniformly to one of the
adjacent cells.
21
22222222IMproves Yourself
Hybrid random direction modelHybrid random direction model
Motion of a node is divided into trips. In a trip, node chooses a direction in [0,360] and moves a distance
Speed of movement (for scaling reasons).
n)
1(
nVn
The average neighborhood size scales as )1
(n
Vn
23232323IMproves Yourself
Discrete random direction model.Discrete random direction model.
Divide the square into cells of area tours of size Time divided into equal duration slots At the beginning of a slot, a node jumps uniformly to an adjacent
cell. During a slot, the node chooses a start and end point uniformly
inside the cell, and moves from start to end. Velocity of motion is made inversely proportional to distance.
2n 21
n
)( 21 n
nn
24242424IMproves Yourself
Brownian motion mobilityBrownian motion mobility
For , the discretized random direction model degenerates to the random walk discrete equivalent of a Brownian motion with variance
2
1
n
1
25252525IMproves Yourself
Critical delay and 2-hop delayCritical delay and 2-hop delay
26262626IMproves Yourself
Definition of critical delayDefinition of critical delay
We know that in the static node case, per node capacity is . Capacity achieving scheme is the multi-hop relaying scheme of Gupta & Kumar.
If mobility is allowed, the two-hop relaying strategy achieves per node capacity of [Grossglauser & Tse]
The two hop relaying strategy has an average delay of , under most mobility models.
Mobility increase capacity at the expense of delay.
)1
(n
)1(
)(n
27272727IMproves Yourself
Definition of critical delay (conti)Definition of critical delay (conti)
Suppose we impose the constraint that the average delay can not exceed .
Under this constraint, relaying strategy that use mobility will achieve a capacity , somewhere between and
For some critical delay bound , this capacity will be equal to capacity of static node networks.
Below this critical delay , there is no benefit from using mobility based relaying.
)(nt
)(nct )1( )1
(n
)(ntc )(nctc
28282828IMproves Yourself
An illustration of critical delayAn illustration of critical delay
Maximum average delay
Capacity
Two hop delay
[Grossglauser & Tse]
[Gupta & Kumar]
Critical delay
Critical delay is the minimum delay that must be tolerated
2 hop relaying scheme has been shown to incur an average delay of about under many different mobility models
29292929IMproves Yourself
More on the critical delayMore on the critical delay
It depends on the mobility mode. It provides a basic to compare mobility model. If mobility model A has lower critical delay than mobility
model B , then A provides more leeway to achieve capacity gains from mobility than B.
Critical delay also depends on what scheduling strategies are allowed.
30303030IMproves Yourself
Critical Delay and 2-hops Delay Under Various Mobility Models
Critical Delay and 2-hops Delay Under Various Mobility Models
31313131IMproves Yourself
Lower bound on critical delay for hybrid random walk models
Lower bound on critical delay for hybrid random walk models
Obtain a value such that if average delay is below this value than (on average) packets travel a constant distance using wireless transmissions before reaching their destinations. For the hybrid random walk model , this value is
Show that if packets are on average relayed over constant distance using wireless transmission, this results in a throughput of ,with the protocol model of the interference.
Thus, the critical delay can not be any lower than this value.
n
n
log
2
)1
(n
32323232IMproves Yourself
Lower bound on critical delay for hybrid random walk models (cont)
Lower bound on critical delay for hybrid random walk models (cont)
Step1: Establish a lower bound on the first exit time from a disc of radius
Step2: If average delay is smaller than , than packets must on average be relayed over a distance no smaller than Pigeonhole argument Exit lemma Union Bound Motion arguments for successful relaying.
81
2
24
1
,
4)
log1024(
nn
nP E
n
nf
log2048
20
2400
10 f
100f
33333333IMproves Yourself
Upper bound on critical delay for hybrid random walk models
Upper bound on critical delay for hybrid random walk models
Develop a scheduling and relaying scheme that provides a throughput of while incurring a delay of
Consider a scheme where relay node transfer the packet to destination when it is in the same cell as destination
Delay=(approx) time for delay node to move into destination node’s cell. Packet arrivals are independent of mobility delay is the same as mean first
hitting time on a torus of size This first hitting time=
)1(n
)log( 2 nn
nn
)log( 2 nn
34343434IMproves Yourself
Upper bound on critical delay for hybrid random walk models (conti)
Upper bound on critical delay for hybrid random walk models (conti)
With this strategy, multi-hop relaying is only used once we reach the destination’s cell, ie., at most distance
Each hop travels a distance Throughput loss from multihop relaying = Since each wireless transmission travels ,nodes within this
range must stay silent. An additional throughput loss of Combining the two, throughput=
)1
( n
)log
(n
n
)log
(2/1
n
nO
)log
(n
n
)(log n
)1
()log
(2
1
nn
n
35353535IMproves Yourself
Discussion on hybrid random walk modelsDiscussion on hybrid random walk models
As increases, the critical delay increases, thereby shrinking the delay-capacity trade-off region.
Two extreme cases: i.i.d model: when the static node capacity can be achieved even with a
constant delay constraint. Random walk model: where delay on the order of is required to
achieve the static node capacity.
)log
(n
n
36363636IMproves Yourself
Lower bound on critical delay for hybrid discretized random direction models
Lower bound on critical delay for hybrid discretized random direction models
Same approach as before to obtain lower bound on critical delay as
Step1: derive a lower bound on exit time from a disc of radius 8 under the random direction model
Step2: If average delay is smaller than packets must on average be relayed over a distance on smaller than
)log
(2
1
n
n
2
21
41
,
4)
log768(
nn
CnP E
n
nCf
log2048
2
1
0
100f
37373737IMproves Yourself
Upper bound on critical delay for hybrid discretized random direction models
Upper bound on critical delay for hybrid discretized random direction models
Same strategy as before Replicate and give to relay node Relay node hands off to destination when it is in the cell of the destination
Can obtain a throughput of with a delay of Provides an upper bound on critical delay for discreted random
direction model.
)1
(n
)log( 2
1
nn
38383838IMproves Yourself
DiscussionDiscussion
39393939IMproves Yourself
Discussion : Characteristic path lengthDiscussion : Characteristic path length
Critical delay seems to be inversely proportion to characteristic path length of a mobility model.
Characteristic path length is the average distance traveled before changing direction under the model.
For example, with hybrid discretized random ditection model, characteristic path length is and the critical delay is n
2
1
n
40404040IMproves Yourself
Discussion : Characteristic path length(cont)
Discussion : Characteristic path length(cont)
Thus, a scenario with nodes moving long distance before changing direction provides more opportunities to harness delay-capacity trade-off, e.g., random way point model vs. Brownian model.
41414141IMproves Yourself
ConclusionConclusion
42424242IMproves Yourself
Conclusion Conclusion
Motivate capacity-delay tradeoff in MANET ( Mobile Ad-hoc NETworks ) .
Define critical delay to compare capacity-delay tradeoff region across mobility models.
Define a parameterized set of hybrid random walk models and hybrid random direction models that exhibit continuous critical delay behavior from minimum possible to maximum possible.
43434343IMproves Yourself
Q&AQ&A
感謝各位聆聽感謝各位聆聽Thanks for your ListeningThanks for your Listening
Q&AQ&A
感謝各位聆聽感謝各位聆聽Thanks for your ListeningThanks for your Listening